10-generation evolution of prompt + LoRA-strength + sampling to maximize Sadness. Fitness = mean over 8 candidates. Best evolved 0.517 vs prompt-without-LoRA 0.000 (LoRA lift +0.517) vs neutral 0.000; winning λ≈1.25.
Method & reproducibility — how the evolution ran
Base model:laion/moss-tts-local-transformer-4.55b-voice-acting-v2 · LoRA: the matching v3 emotion LoRA (TTS-AGI/moss-emotion-loras-v3, rank 32), applied via PEFT adapter scaling at strength λ. Algorithm: genetic search — population 16, 10 generations, elitism (top 6 carried over). Each generation: every genome is generated best-of-8 and its fitness = the mean target-emotion strength over those 8 candidates (EmoNet score for the emotion, minus 0.25·max(0, WER−0.6) as a light intelligibility guard). Offspring = 60% crossover of two elites + mutation, 40% mutation of one elite. Genome (what evolves): the GENERAL voice description, the SCRIPT delivery cue, the carrier sentence, the LoRA strength λ (seeded 0.75–1.5, mutated within 0.4–2.0), and the sampling params (audio temperature, top_p, top_k). Mutation operators: 35% regenerate the description (new intensity + manner adjectives), 20% swap the carrier sentence, 23% jitter λ by ±0.25/0.5, 12% jitter temperature ±0.1, 10% change top_k. LoRA on vs off: the two LoRA OFF cards are baselines generated with the adapter disabled (λ=0) — a plain neutral prompt, and the best steering prompt without the LoRA. Every LoRA ON hall-of-fame recipe below was generated with the LoRA active at the evolved λ shown on the card. Seed is derived from the emotion name (deterministic).
Baselines — LoRA disabled
NEUTRAL promptLoRA OFF (λ=0)
mean-emo 0.000 · best 0.00 · blend 0.21 · genu 0.05 · WER 0.47
STEER promptLoRA OFF (λ=0)
mean-emo 0.000 · best -0.00 · blend 0.41 · genu 0.09 · WER 0.34
GENERAL: A voice viscerally expressing sadness, a soft, grief-worn voice, heavy and slow, on the edge of tears, in every breath.
SCRIPT: (quietly, grief-stricken, near tears) "The meeting is scheduled for three o'clock on Tuesday afternoon."
Hall of fame — top 8 evolved recipes (all generated WITH the emotion LoRA active)
#1LoRA ON · λ=1.25 (125%)
mean-emo 0.517 · best 0.69 · blend 0.74 · genu 0.35 · WER 0.54
GENERAL: A voice powerfully expressing sadness, a soft, grief-worn voice, heavy and slow, on the edge of tears, impossible to hide.
SCRIPT: (quietly, grief-stricken, near tears) "After everything, it comes to this."
mean-emo 0.531 · best 0.63 · blend 0.51 · genu 0.34 · WER 0.80
GENERAL: A voice powerfully expressing sadness, a soft, grief-worn voice, heavy and slow, on the edge of tears, impossible to hide.
SCRIPT: (quietly, grief-stricken, near tears) "I feel it, all of it, right now."
mean-emo 0.513 · best 0.58 · blend 0.66 · genu 0.34 · WER 0.73
GENERAL: A voice intensely expressing sadness, a soft, grief-worn voice, heavy and slow, on the edge of tears, with every word.
SCRIPT: (quietly, grief-stricken, near tears) "After everything, it comes to this."
mean-emo 0.532 · best 0.71 · blend 0.47 · genu 0.32 · WER 0.81
GENERAL: A voice to the extreme expressing sadness, a soft, grief-worn voice, heavy and slow, on the edge of tears, in every breath.
SCRIPT: (quietly, grief-stricken, near tears) "After everything, it comes to this."
mean-emo 0.518 · best 0.66 · blend 0.59 · genu 0.36 · WER 0.78
GENERAL: A voice powerfully expressing sadness, a soft, grief-worn voice, heavy and slow, on the edge of tears, building and building.
SCRIPT: (quietly, grief-stricken, near tears) "I feel it, all of it, right now."
mean-emo 0.470 · best 0.59 · blend 0.65 · genu 0.36 · WER 0.58
GENERAL: A voice deeply expressing sadness, a soft, grief-worn voice, heavy and slow, on the edge of tears, pouring out.
SCRIPT: (quietly, grief-stricken, near tears) "After everything, it comes to this."
mean-emo 0.468 · best 0.59 · blend 0.71 · genu 0.29 · WER 0.60
GENERAL: A voice to the extreme expressing sadness, a soft, grief-worn voice, heavy and slow, on the edge of tears, with every word.
SCRIPT: (quietly, grief-stricken, near tears) "After everything, it comes to this."
mean-emo 0.528 · best 0.68 · blend 0.53 · genu 0.34 · WER 0.84
GENERAL: A voice powerfully expressing sadness, a soft, grief-worn voice, heavy and slow, on the edge of tears, raw and unfiltered.
SCRIPT: (quietly, grief-stricken, near tears) "I feel it, all of it, right now."
